(249m) Understanding Rare Safety and Reliability Events Using Transition Path Sampling

In the chemical manufacturing industry, processes and their control systems are typically well-designed to mitigate events that have the potential to lead to plant shutdowns or safety events â?? those that often result in human health losses or environmental impacts. Clearly, there is strong motivation to understand how these events develop and propagate. Effective operator training, safety system design, and safety analysis, all benefit from a full understanding of such events. A major challenge in the study of events that propagate to process shutdown or safety incidents is their sparsity â?? typically these events occur so rarely that statistical techniques alone are incapable of describing and characterizing them â?? especially when they have not yet occurred. Simulation of these events could be useful to understand them, however, a daunting computational challenge exists. Typical rare events occur on the order of years or decades apart, yet the events occur within minutes or hours. Thus, the bulk of the computational effort in simulating rare events is allocated to normal operation, making the event computationally infeasible to simulate with meaningful frequency.

A rare-event sampling technique, Transition Path Sampling (TPS), has been developed by the molecular dynamics community. While the time and length scales between molecular dynamics and process dynamics differ greatly, the ratios of the time of the rare events and the waiting time between them are similar. This Monte-Carlo based technique relies on the simulation of perturbed rare-event trajectories â?? an initial rare-event trajectory is randomly modified such that large numbers of trajectories are generated. Clusters of rare safety-event trajectories can be the basis for alarm and safety-system design, assuring that TPS-generated clusters are preventable. Important modifications to the TPS technique are needed to apply it to process dynamics. The backwards integration, a key attribute of TPS, is not possible for most process simulations â?? instead a boundary-value optimization technique is used. Furthermore, process models use of the vast amount of process data available for model verification and to estimate the relative likelihood of one trajectory to another. The application of TPS will be demonstrated using a simple jacketed exothermic CSTR, as well as a more complex air separation unit process. This innovative approach allows for a quantitative rationalization of alarm and safety systems to prevent rare, yet serious, safety events.